Microsoft Corp. today is announcing at its annual Build conference a broad range of enhancements across its data platforms aimed at simplifying artificial intelligence application development.

The updates span Microsoft Fabric, Azure Cosmos DB, SQL Server, Power BI, PostgreSQL and integrations with Databricks Inc.’s and SAP SE’s services on the Azure cloud. The aim is to offer developers a more cohesive and efficient toolkit for building intelligent applications.

Cosmos DB comes to Fabric

Microsoft Fabric, a data platform intended to unify all data and analytics services, now includes support for semistructured data through the preview of Azure Cosmos DB, a NoSQL and vector database. This allows developers to manage unstructured and semistructured data types such as documents, text and emails for analytics.

Cosmos DB is known for its scalability, reliability and support of both SQL and NoSQL. Data is automatically accessible in OneLake, the unified data lake storage system built into Microsoft Fabric. That enables real-time analytics scenarios like sentiment analysis in chat applications.

Fabric Real-Time Intelligence for streaming data now includes a digital twin builder in preview. This no-code/low-code tool helps developers model and manage virtual replicas of physical and logical entities, Microsoft said. It’s designed for what-if analysis and process modeling in scenarios such as manufacturing, logistics and customer profiling.

Microsoft is also expanding the use of Azure Cosmos DB within the Azure AI Foundry unified AI development platform to support AI agents that remember and resume prior user interactions. This thread storage capability, now generally available, allows AI agents to maintain contextual continuity in conversations, which is a critical component for natural user experiences.

A new full-screen Copilot in Power BI enables users to query data conversationally. Copilot can search across multiple reports and semantic models, integrating directly into Microsoft Teams and Microsoft 365. Fabric data agents created in Copilot Studio can now be embedded in Microsoft 365 and Teams, automating routine tasks and enabling deeper interaction with enterprise data.

Vector embeddings on SQL Server

Keeping the 35-year-old SQL Server relevant in the AI age, the 2025 edition, now in public preview, has been expanded to include vector database functionality. It supports AI applications through native vector search, local or cloud-based AI model interfaces and integration with frameworks such as LangChain for large language models and Semantic Kernel for AI agents. It also supports real-time data mirroring into Fabric and deployment through Azure Arc.

PostgreSQL on Azure is also getting significant upgrades to help developers extract more meaningful context from relational data. A new VS Code extension, integrated with GitHub Copilot, streamlines database development with AI assistance for query writing and schema design. DiskANN, a high-performance vector indexing algorithm, is now generally available in Azure Database for PostgreSQL, offering a faster alternative to pgvector.

Semantic operators powered by large language models are also in preview, enabling GenAI reasoning directly within PostgreSQL.

Databricks and SAP Integration

Microsoft also plans to tighten its integration with SAP’s Business Data Cloud and SAP Databricks on Azure early this summer. The platform will allow developers to build AI applications on top of SAP business data using Microsoft’s first-party Databricks service. This setup offers scalability and security while maintaining tight integration with Azure.

Mirroring capabilities now allow Azure Databricks Unity Catalog tables to be synced in real-time with OneLake. Developers can reference Databricks tables across services with minimal setup, streamlining data access and management across cloud environments.

Azure AI Foundry agents can now execute Databricks jobs and use Databricks’ AI/BI Genie conversational interface for data retrieval and contextual response generation.

Photo: Unsplash